A Cnn-Gru Approach to Capture Time-Frequency Pattern Interdependence for Snore Sound Classification

2018 ◽  
Author(s):  
Jianhong Wang ◽  
Harald Stromfeli ◽  
Bjorn W. Schuller
Keyword(s):  
Frequency Pattern ◽  
Time Frequency ◽  
Capture Time ◽  
Sound Classification

An automatic means to discriminate between earthquakes and quarry blasts

1990 ◽  
Vol 80 (6B) ◽  
pp. 2143-2160
Author(s):  
Michael A. H. Hedlin ◽  
J. Bernard Minster ◽  
John A. Orcutt
Keyword(s):  
Fourier Transform ◽  
Spectral Features ◽  
Single Event ◽  
Vertical Array ◽  
Signal To Noise ◽  
Spectral Character ◽  
Frequency Pattern ◽  
Time Frequency ◽  
Cepstral Analysis ◽  
Double Fourier Transform

Abstract In this article we discuss our efforts to use the NORESS array to discriminate between regional earthquakes and ripple-fired quarry blasts (events that involve a number of subexplosions closely grouped in space and time). The method we describe is an extension of the time versus frequency “pattern-based” discriminant proposed by Hedlin et al. (1989b). At the heart of the discriminant is the observation that ripple-fired events tend to give rise to coda dominated by prominent spectral features that are independent of time and periodic in frequency. This spectral character is generally absent from the coda produced by earthquakes and “single-event” explosions. The discriminant originally proposed by Hedlin et al. (1989b) used data collected at 250 sec−1 by single sensors in the 1987 NRDC network in Kazakhstan, U.S.S.R. We have found that despite the relatively low digitization rate provide by the NORESS array (40 sec−1) we have had good success in our efforts to discriminate between earthquakes and quarry blasts by stacking all vertical array channels to improve signal-to-noise ratios. We describe our efforts to automate the method, so that visual pattern recognition is not required, and to make it less susceptible to spurious time-independent spectral features not originating at the source. In essence, we compute a Fourier transform of the time-frequency matrix and examine the power levels representing energy that is periodic in frequency and independent of time. Since a double Fourier transform is involved, our method can be considered as an extension of “cepstral” analysis (Tribolet, 1979). We have found, however, that our approach is superior since it is cognizant of the time independence of the spectral features of interest. We use earthquakes to define what cepstral power is to be expected in the absence of ripple firing and search for events that violate this limit. The assessment of the likelihood that ripple firing occurred at the source is made automatically by the computer and is based on the extent to which the limit is violated.

scholarly journals Individual Stability of Pain- and Touch-Related Neuronal Gamma Oscillations

2021 ◽  
Author(s):  
Elia Valentini ◽  
Alina Shindy ◽  
Viktor Witkovsky ◽  
Anne Stankewitz ◽  
Enrico Schulz
Keyword(s):  
Gamma Oscillations ◽  
Cortical Processing ◽  
Neuronal Oscillations ◽  
Similar Time ◽  
Frequency Pattern ◽  
Research Results ◽  
Time Frequency ◽  
Gamma Range ◽  
Broad Variety ◽  
Diverse Individual

Background: The processing of brief pain and touch stimuli has been associated with an increase of neuronal oscillations in the gamma range (40-90 Hz). However, some studies report divergent gamma effects across single participants. Methods: In two repeated sessions we recorded gamma responses to pain and touch stimuli using EEG. Individual gamma responses were extracted from EEG channels and from ICA components that contain a strong gamma amplitude. Results: We observed gamma responses in the majority of the participants. If present, gamma synchronisation was always bound to a component that contained a laser-evoked response. We found a broad variety of individual cortical processing: some participants showed a clear gamma effect, others did not exhibit any gamma. For both modalities, the effect was reproducible between sessions. In addition, participants with a strong gamma response showed a similar time-frequency pattern across sessions. Conclusions: Our results indicate that current measures of reproducibility of research results do not reflect the complex reality of the diverse individual processing pattern of applied pain and touch. The present findings raise the question of whether we would find similar quantitatively different processing patterns in other domains in neuroscience: group results would be replicable but the overall effect is driven by a subgroup of the participants.

Instantaneous Engine Speed Time-Frequency Analysis for On-Board Misfire Detection and Cylinder Isolation in a V12 High Performance Engine

2002 ◽  
Author(s):  
Fabrizio Ponti
Keyword(s):  
Frequency Analysis ◽  
Engine Speed ◽  
Test Bench ◽  
Specific Pattern ◽  
Operating Conditions ◽  
False Alarms ◽  
Frequency Pattern ◽  
Time Frequency Analysis ◽  
Time Frequency ◽  
Misfire Detection

The diagnosis of a misfire event and the isolation of the cylinder in which the misfire took place is enforced by the On Board Diagnostics (OBD) requirements over the whole operating range for all the vehicles whatever the configuration of the engine they mount. This task is particularly challenging for engines with a high number of cylinders and for engine operating conditions that are characterized by high engine speed and low load. This is why much research has been devoted to this topic in recent years, developing different detection methodologies based on signals such as instantaneous engine speed, exhaust pressure, etc., both in time and frequency domains. This paper presents the development and the validation of a methodology for misfire detection based on the time-frequency analysis of the instantaneous engine speed signal. This signal contains information related to the misfire event, since a misfire occurrence is characterized by a sudden engine speed decrease and a subsequent damped torsional vibration. The identification of a specific pattern in the instantaneous engine speed frequency content, characteristic of the system under study, allows performing the desired misfire detection and cylinder isolation. Particular attention has been devoted in designing the methodology in order to avoid the possibility of false alarms caused by the excitation of this frequency pattern independently from a misfire occurrence. Although the time-frequency analysis is usually considered a time consuming operation and is not associated to on-board application, the methodology here proposed has been properly modified and simplified in order to obtain the quickness required for its use directly on-board a vehicle. Experimental tests have been performed on a 5.7 liter V12 spark ignited engine, with the engine mounted on-board a vehicle. The frequency pattern identified is not the same that could be observed when running the engine on a test bench, because of the different stiffness that the connection between the engine and the load presents in the two cases. This makes impossible to set-up the methodology here proposed only on a test bench, without running tests on the vehicle.

Instantaneous Engine Speed Time-Frequency Analysis for Onboard Misfire Detection and Cylinder Isolation in a V12 High-Performance Engine

2008 ◽  
Vol 130 (1) ◽  
Author(s):  
Fabrizio Ponti
Keyword(s):  
Frequency Analysis ◽  
Engine Speed ◽  
Test Bench ◽  
Operating Conditions ◽  
False Alarms ◽  
Frequency Pattern ◽  
Time Frequency Analysis ◽  
Time Frequency ◽  
Resonance Frequencies ◽  
Misfire Detection

The diagnosis of a misfire event and the isolation of the cylinder in which the misfire took place is enforced by the onboard diagnostics (OBD) requirements over the whole operating range for all the vehicles, whatever the configuration of the engine they mount. This task is particularly challenging for engines with a high number of cylinders and for engine operating conditions that are characterized by high engine speed and low load. This is why much research has been devoted to this topic in recent years, developing different detection methodologies based on signals such as instantaneous engine speed, exhaust pressure, etc., both in time and frequency domains. This paper presents the development and the validation of a methodology for misfire detection based on the time-frequency analysis of the instantaneous engine speed signal. This signal contains information related to the misfire event, since a misfire occurrence is characterized by a sudden engine speed decrease and a subsequent damped torsional vibration. The identification of a specific pattern in the instantaneous engine speed frequency content, characteristic of the system under study, allows performing the desired misfire detection and cylinder isolation. Particular attention has been devoted to designing the methodology in order to avoid the possibility of false alarms caused by the excitation of this frequency pattern independently from a misfire occurrence. Although the time-frequency analysis is usually considered a time-consuming operation and not associated to onboard application, the methodology proposed here has been properly modified and simplified in order to obtain the quickness required for its use directly onboard a vehicle. Experimental tests have been performed on a 5.7l V12 spark-ignited engine run onboard a vehicle. The frequency characteristic of the engine-vehicle system is not the same that could be observed when running the engine on a test bench, because of the different inertia and stiffness that the connection between the engine and the load presents in the two cases. This makes it impossible to test and validate the methodology proposed here only on a test bench, without running tests on the vehicle. Nevertheless, the knowledge of the mechanical design of the engine and driveline gives the possibility of determining the resonance frequencies of the system (the lowest one is always the most important for this work) before running tests on the vehicle. This allows saving time and reducing costs in developing the proposed approach.

scholarly journals An Interpretable Deep Learning Model for Automatic Sound Classification

Electronics ◽  
2021 ◽  
Vol 10 (7) ◽  
pp. 850
Author(s):  
Pablo Zinemanas ◽  
Martín Rocamora ◽  
Marius Miron ◽  
Frederic Font ◽  
Xavier Serra
Keyword(s):  
Deep Learning ◽  
Web Application ◽  
Domain Knowledge ◽  
Learning Model ◽  
Unintended Effects ◽  
Learning Models ◽  
Time Frequency ◽  
Sound Classification ◽  
Proposed Model ◽  
Deep Learning Model

Deep learning models have improved cutting-edge technologies in many research areas, but their black-box structure makes it difficult to understand their inner workings and the rationale behind their predictions. This may lead to unintended effects, such as being susceptible to adversarial attacks or the reinforcement of biases. There is still a lack of research in the audio domain, despite the increasing interest in developing deep learning models that provide explanations of their decisions. To reduce this gap, we propose a novel interpretable deep learning model for automatic sound classification, which explains its predictions based on the similarity of the input to a set of learned prototypes in a latent space. We leverage domain knowledge by designing a frequency-dependent similarity measure and by considering different time-frequency resolutions in the feature space. The proposed model achieves results that are comparable to that of the state-of-the-art methods in three different sound classification tasks involving speech, music, and environmental audio. In addition, we present two automatic methods to prune the proposed model that exploit its interpretability. Our system is open source and it is accompanied by a web application for the manual editing of the model, which allows for a human-in-the-loop debugging approach.

Sign in / Sign up

Export Citation Format

Share Document